In the crime book I wrote briefly about two extremely similar murder stories, which were the murder of Martha Moxley—a distant but not too-distant cousin of my wife--and the murder of Robin Gilbert.  In both of these cases no one was arrested or prosecuted at the time of the crime, but many years later people were arrested, tried and convicted.   The convicted in both cases were

                1)  Ne’er-do-wells, and

                2)  Teen-aged associates of the dead girl. 

                In both cases the prosecutors convinced the jury that the ne’er-do-well had committed the crime, but failed to convince me.  I think frankly that it is more likely that Martha Moxley and Robin Gilbert were murdered by the same unknown person than that they were murdered by Michael Skakel and David Allen Jones.

                In commenting on this I wrote the following paragraph: 

There should be some way to estimate mathematically the probability that two crimes are linked.   We have here two young girls of essentially the same age, living in safe neighborhoods, who died under extremely similar circumstances, separated by neither 180 days nor 180 miles.   Both girls’ clothes were torn open, but they were not raped, suggesting an attacker who might be impotent.   Is this enough to suggest that the crimes are linked?   I don’t know.   From prison, both Skakel and Jones continue to deny that they had any involvement in the crimes. 

My purpose today was to pick up this question, "How could we estimate mathematically the probability that two crimes are linked?"

It’s obvious how we start that process.   We start that process by "coding" the crime so that it has a mathematical signature.    We "code" the crime by asking a series of specific questions, like "Was the victim Male or Female?"

In these two cases the victim was Female, but let’s work with a field of a dozen or so cases.   I’ll list ten cases from Popular Crime and throw in two others.   We will work with the murders of Dr. George Parkman, Andrew Borden, Abby Borden, Paul Stine, Cheri Jo Bates,

Louis Masgay, John Davies, Gregg Smart, Martha Moxley, Robin Gilbert, Robbin Brandley and Michelle O’Keefe.   Alphabetically:

                I am going to have to break in here and write little two-sentence synopses of these twelve cases:

                Cheri Jo Bates was a very cute college girl, a student at Riverside City College on the south side of LA, who was murdered on October 30, 1966.   The killer wrote a long letter to police, claiming that he would be murdering many more.   The case is often linked to the Zodiac, who terrorized San Francisco three years later. 

                Abby and Andrew Borden were the parents of Lizzie Borden.

                Robbin Brandley was a student at Saddleback College in Mission Viejo, California (Orange County).   On January 18, 1986, she worked as a volunteer at a Jazz Concert on the campus.  Walking back to her car after the concert after dark she was accosted in the parking lot, and stabbed 41 times.

                In 1997 Andrew Urdiales confessed to the crime.   Urdiales became a serial murderer; this was his first murder.   However, Brandley’s parents feel strongly that others were involved in her murder, as well, and they have conducted a 25-year war against the Orange County Sherriff’s department, trying to force the filing of additional charges.

                John Davies was a 15-year-old boy who disappeared from his home in Belmont, California, in 1981.   Davies was close friends with a serial murderer named Jon Dunkle, who is believed to have killed Davies for unknown reasons.

                Robin Gilbert was a 14-year-old girl living in a very safe neighborhood in Reading, Massachusetts.   On July 2, 1975, she snuck out of her house after bedtime, probably to meet a boy and smoke a couple of cigarettes.  Her body was found the next morning, buried under leaves and debris at the end of 100 feet of drag marks. 

                Louis Masgay was a businessman who was so unwise as to take a large wad of cash to a meeting with Richard Kuklinski.   When his body was discovered two years later, his insides were frozen, earning Kuklinski the nickname The Ice Man.   Kuklinski talked at length about his crimes in three HBO documentaries, making him one of the most famous murderers of his generation.   (I don’t remember who it was, but I remember that one of the posters on this site lived down the street from Kuklinski at the time.)

                Martha Moxley was a 15-year-old girl living in Greenwich, Connecticut; the man convicted of her murder, Michael Skakel, was a member of the Kennedy family, which makes the crime quite famous.

                Michelle O’Keefe was an actress and a student at Antelope Valley College north of LA.  On February 22, 2000, she appeared in a Kid Rock video which required that she dress provocatively (she can still be seen in the video).   Returning to her car in a parking lot about 9:30 that evening, she was shot four times for no apparent reason.    NBC did a Dateline episode about the case; google The Girl with the Blue Mustang if you’re interested. 

                George Parkman was the victim in one of the most famous cases in 19^th century America, known as The Harvard Murder.

                Greg Smart was the murdered husband of Pam Smart, a "teacher"—not really a teacher—who seduced a 15-year-old boy and persuaded him to kill hubby.   The case was fictionalized in the quite wonderful movie To Die For, starring Nicole Kidman.

                Paul Stine was a cab driver who was murdered by the Zodiac in San Francisco in 1969. 

                OK, the next thing that we will need to code for is the age of the victim.    Here we will have to make our first non-obvious choice.    We could, on the one hand, code the person’s age simply by their calendar age at the time of their death, or we could "group" the ages to represent the stages of life:

                1)  Infancy,

                2)  Pre-pubescent Childhood,

                3)  Teenagers (meaning kids who have gone through puberty but are not yet at an age where they would ordinarily have left home,

                4)  College age,

                5)  Young adult (ages 23-34),

                6)  Prime of life (Ages 35 to 55),

                7)  Post-prime (Ages 56 to 75),

                8)  Really old.

                I seem to remember Shakespeare declaring that there were seven stages to life, not eight, but. . .what did Shakespeare know?  I don’t see that there is any real advantage to making the age specific.   If we think about an unknown killer. . .does the killer either know or care exactly when the victim was born?   A criminal preys on a certain type of victim, often a certain age range.    He is not too likely to care whether his victim is 16 or 17.    It doesn’t seem relevant.   I’m going the other way.

                At this point, then, we have 16 possible codes (two times eight). There are about 15,000 murders in the US every year, so that’s about 1,000 victims per code, more or less. ..obviously no basis on which any two murders can be linked.    Let’s add the race/origin of the victim—A for African American, C for Caucasian, O for Oriental, N for Native American, H for Hispanic.    All of these were plain white people:

                So we don’t gain anything there, but in theory we now have 80 codes, rather than 16. Now we get to the tricky stuff.   Where was the victim found?

                Ham radio operators use a six-digit code called the Maidenhead Grid System to represent land spaces.    DM13hx, for example, represents Riverside, California, where Cheri Jo Bates was murdered.   This seems to be a near-perfect system for what we need, so let’s enter the codes for the places were these murders were committed:

                Some of those may not be exactly right, but. . .that’s not really the point.   On the first element "C" corresponds to the far west coast of the US, "D" to the western half of the US, generally, "E" to the eastern half of the US, generally, and "F" to the east coast.   "M" corresponds to the southern half of the US, generally, and "N" to the northern half, although there are small parts of the US which are in "L" and "O".   A six-element code represents an area about 5 miles wide and 2.5 miles high, and the distance between any two areas can be very easily calculated.   Fenway Park is in FN42ki, and you can see that the murders of Abby and Andrew Borden (FN41), Robin Gilbert (FN42), George Parkman (FN42) and Gregg Smart (FN42) all occurred within 50 or 60 miles of Fenway.  

                OK, at this point we have more codes than murders, and we’re just getting starting.   The next thing we might ask is whether the victim was found inside a dwelling, inside a building/not a dwelling, in a vehicle, in an urban area but not inside a building, or in a rural area:

                B—Inside a building not used as a dwelling

                D—Inside a dwelling

                R—In a rural area/not inside a vehicle or building

                U—In an urban area, but not in a building or a vehicle

                V—Inside a vehicle

                And we will need a sixth code there, which is "X", meaning that the body was never found.   I believe that the body of John Davies was never found:

                Ignoring the six-digit location elements we now have 480 possible codes.   The location given for Davies, whose body was never found, is the location at which he was last seen, inside his own home in Belmont.

                Next we could ask "was the body found at the location where the person was killed?", yes (1) or no (2)?    It is, in almost all cases, very easy to tell.    Homicide investigators know whether a body was killed at the site or was dumped there, in most cases, the moment they see the body.  The only time they don’t know is when the body isn’t found for several weeks or sometimes if the person was killed in a bloodless way.

                Since Davies’ body was not found, we enter that as "3". We are up to 960 possible codes now, not including the location code.    The next thing we could ask about is the cause of death.

                A—Asphyxiation (Suffocated)

                B—Blunt Force Trauma (Bludgeoned)  

                D--Drowning

                G—Gunshot Wound (single)

                M—Multiple Gunshots

                P—Poisoned

                S--Stabbed

                T—Strangled

                X—Other

                Z—More than one cause of death

                There are a million possible causes of death; I have heard of murder cases in which the victim was starved, dehydrated (denied water), or killed by carbon monoxide poisoning (I’m not sure whether that would technically be poisoning or asphyxiation.)  I’ve heard of people being murdered by tampering with the brakes on their car (which I suppose would technically be blunt force trauma) and I’ve heard of people being killed with poisonous snakes and other animals.  I’ve heard of people being killed by lions and tigers.   I’ve never heard of anyone being murdered by bee stings, but I would suppose that it must have happened.  I have heard of people having heart attacks while being held captive, which is murder although the "weapon" is the person’s own body.    I’ve heard of people being run over with trucks, electrocuted, buried alive and having air injected into their veins.

                I don’t think it would be productive toward our general purposes to make a separate category for each of these events; I think it is better to break it off with "other" and we’ve probably got 98% of the cases in the other nine categories. 

                With these ten we now have 9,600 "death codes", meaning there would be about one to two murders per code per year in the United States.   Let us move on now to the instrument of death.   The instrument of death is tied closely to the cause of death; i.e, we would not say that a person was shot with a rope or strangled with a shotgun or poisoned with a hammer.   The causes and instruments of death from Asphyxiation (A) could be summarized this way:

In one of the last episodes of the Sopranos, Tony Soprano kills a longtime friend by holding his nose closed after a car accident has left him gasping for breath.   Such things do happen; one could not easily make a list of all the instruments that could be used to suffocate someone, and it is often quite difficult to know how a person was suffocated.

                One might assume that a veteran detective would know immediately whether a person was shot or stabbed or bludgeoned, and while this is often true it is not always true.   Stab wounds can look like gun shots; gun shots can look like blunt force trauma.  We could group the deaths by Blunt Force Trauma (B) this way:

                Are people killed with brass knuckles anymore?   It used to be that muggers and thugs would wear brass knuckles, and there were a fair number of murders committed with brass knuckles, but I don’t know that that happens anymore.  

                The distinction between being killed by a single blow and a series of blows is quite important, because if a person is killed by a single blow, that can be an accident or the result of a fight in which no one intended to seriously injure anyone, whereas if a person is killed by a series of blows, it may be assumed that the assailant intended for the assault to end in death or serious injury.  The full list of causes and instruments of death will be presented at the end of today’s article; I don’t want to do that here because it is a long list and it drags us a long way away from our discussion.    But there is an array of 56 categories of causes and instruments of death, which gives us about 54,000 possible codes to represent a murder victim.  Let us add the codes to the twelve cases that we are tracking:

                I have entered the cause of death for Robin Gilbert as "X2", meaning the cause of death is unknown.   The Medical Examiner ruled at the time of her death that Gilbert—a 14-year-old girl with no history of heart trouble--had died of heart disease, a conclusion that can truly be described as bizarre.   Twenty years later her body was exhumed; a new medical examiner ruled that Robin Gilbert did not die of heart disease, and that her death was a homicide, which was obvious anyway.   At his trial, David Allen Jones was accused of strangling the girl; however, to the best of my knowledge there is no medical evidence that she was strangled (no broken hyoid bone, for example.)   I have entered the cause of death as unknown because I am trying to code the death as the detective who investigated the case would have seen it.   The cause of death was not evident from the body. 

                The next question I would ask is as to the apparent motive for the crime.   Of course, in some cases the motive for a murder is not apparent from the scene of the crime, but in most cases it is:

                A—Extreme Anger

                R--Robbery

                S—Sexual Assault

                X—No motive apparent

                If an attractive young woman is murdered and her clothes are torn, one can presume that’s a sexual assault.   If a convenience store clerk is shot, one presumes that’s a robbery.   If a person sitting in a parked car is shot by unknown persons, you don’t know what that is:

                Now that I look at those, many of these "motive codes" are borderline calls.    Cheri Jo Bates was a very attractive young woman and we know for certain that the motive was sexual, because the killer wrote a letter to the police, but I don’t know that that was apparent from the crime scene.  The Bordens were killed for no apparent reason, but with such redundant and unnecessary violence that it is easy to say that’s a rage killing, an anger killing, and it would be difficult to say that it was anything else.  Robbin Brandley was stabbed 41 times, but for no obvious reason.   Michelle O’Keefe was an attractive college girl wearing a tube top that barely contained her assets, but there is no actual evidence that this is why she was killed.   Gregg Smart’s killers faked a robbery.. ..well, they committed an actual robbery to confuse the police, not very successfully.

                Paul Stine was the cab driver murdered by the Zodiac.   San Francisco had had a string of taxicab murders, and police who responded to the scene assumed that this was just another one, and also that the perpetrator would be identified very quickly because he seemed to have made a number of mistakes.   Future events would prove that a) it wasn’t a robbery, and b) that the case would never be solved, but police responding to the scene just assumed it was a robbery. 

                Anyway, this gives us about 216,000 possible codes, not counting the location, and we’ve got four questions to go.   The next question is, "Was the victim engaged in high-risk activity?"

                1)  The victim was clearly not engaged in a high-risk activity at the time of the murder.

                2)  The victim was engaged in an entirely legitimate business or activity which carries a degree of known risk,

                3)  The victim was engaged in a high-risk activity such as prostitution, drug sales or associating with criminal gangs.

                 4)  Unknown.

                At this point you’ve lost track in your mind of what all of them numbers and letters mean, but don’t worry about it; we’ll work on that later.   Three questions to go; we’ve got 850,000 possible codes.   The next question has to do with the date of the murder or, if that is unknown, the estimated date of the person’s disappearance.   As tempted as I am to make up some sort of code here, I’m just going to go with the year and the Julian date.   I’ll list them this time in chronological order:

                For a quite different reason, we also need to enter the time of day of the murder; not the exact time, but the part of the day:

                A—Morning (Dawn until noon)

                B—Afternoon/Evening (Noon until sundown)

                C—Early Night (Sundown until midnight)

                D—Late Night (Midnight until Dawn)

                X—Unknown

                Not counting the specific date and the specific location, we now have about four million possible codes for murders.   The final question will be the most complicated to code of the group, and, since I’m just outlining the idea here, I won’t get too deeply into the details. 

                I’d like to include some way to code "signature" elements of a crime.  What we are looking for here in general is the "signature" of a crime; if a criminal abducts young boys, that is part of his signature, but our system would recognize that pattern already.   I’m speaking now of things that we might miss.  The Zodiac wrote messages at the scenes of some of the crimes; so did Charles Manson’s minions, Jack the Ripper (possibly) and the late William Heirens, a serial murderer from Chicago in the 1940s who just died a couple of weeks ago.    It is uncommon behavior for a murderer, but it is not so uncommon that we wouldn’t recognize it.  

                There are millions of things that a criminal might do that would constitute a recognizable signature.   Our challenge is to generalize about these things in such a way that we can help to recognize patterns, but storing the information in a code of one or at most two digits.  The Boston strangler tied bows around his victims’ necks.  Among the obvious "signatures" are stealing the victims’ jewelry, stealing their socks or some similar item, stealing their purses or wallets, stealing identification or keys, stalking the victim before the attack, cutting phone lines into the house (back when houses had phone lines), tying up victims in an identifiable way (nylon cord, duct tape, electrical cords), contacting police and/or relatives and/or media after the attack, wiping the crime scene clean, breaking out a bathroom window to gain entrance, dismembering the corpse, leaving the body naked in a semi-public place, and placing something from the scene of the crime on the victim’s body.   Let’s make a list of those:

                We can add other "signature elements" to the list later on.   It occurs to me that the list is useful to explain the uniqueness of the JonBenet Ramsey case.   The killer of JonBenet did five things from this list—a, g, k, m, and r.   The usual number of boxes you would check off here in describing a murder would be zero; in the Ramsey case, it is five.   Anyway, our final codes:

Final, for now.    We’ve still got a lot of work to do.

Conclusion for Today

                Up to now, I have sort of been assuming that the reasons for doing this are obvious.    Probably they are obvious to many of you and the rest of you stopped reading several pages ago, but. …just in case anybody asks.

                There is a pervasive need, in the law enforcement process, to know whether two crimes are or are not linked.    When a detective stalls out on a case and is not ready to give up on it, one of the first things he will do is to start looking for related cases.   Once he starts that process, he will find a vast field of unsolved cases, with nothing except intuition to guide him as to which cases may be related. 

                It would be no effort at all, relative to the vast resources of the police network, to maintain a file with the codes of every murder in the United States.    It would be a very small project to educate police about the coding system.   Once that was done, a policeman searching for related cases would be in a vastly better position.   We could make the most similar cases, from anywhere in the United States, jump out from the list like a rapper’s bling.   We could make the process of identifying similar crimes a thousand times more efficient and a dozen times more productive than it is.

                This is merely one situation in which The System needs to know whether cases are or are not related.    A big-city police force is often dealing with 25 or 50 (or more) unsolved cases in a few months, and it is always the case that some of those crimes will look alike.   The police have to make a decision:   are these cases genuinely related, or coincidentally similar? Do we combine these investigations, or do we not?   

                When you suspect that you are dealing with a serial murderer, you need to know.   When you have a serial murderer, you need a task force.  Do we have a serial murderer here, or some coincidences?  

                Michael Skakel was unsuccessfully defended by an excellent defense attorney, Mickey Sherman, who (I believe) is sincerely convinced of Skakel’s innocence.   Sherman needs to know—needed to know, when he was defending Skakel—whether the Robin Gilbert case was or was not a related homicide.   It’s a matter of great urgency that he have the best possible information on that issue.

                When the defense claims that the murder in question was a part of a series of crimes and that evidence of related and similar crimes should be admitted into a trial, the judge has to decide whether it is or is not credible that the cases are related. In the case of Rabbi Neulander, who was convicted of arranging the murder of his wife Carol, there was a very similar murder committed a few miles away about a year later.   Len Jenoff, the "Private Investigator" who actually murdered Carol Neulander, also worked as an "Investigator" in the other case.   Neulander’s defenders wanted to introduce evidence about the other murder.   The judge ruled—incorrectly, in my view—that the cases were not related and that the other murder could not be discussed.

                A jailhouse informant claimed that Jenoff confessed to him about committing the other murder as well.    It is among the scandals of the American judicial system that when a jailhouse snitch claims to have evidence implicating the accused, the evidence is almost always admitted, but when he has evidence tending to exonerate the accused, he is very often not allowed to testify.    But that’s off the topic, which is:  is it most reasonable to believe that these two murders were connected, or is it more reasonable to believe that they were not?

                Vast resources can be (and are) wasted in police investigations when the police mistakenly believe that two crimes are related when in fact they are not.   More significantly, opportunities to solve crimes are missed when police fail to recognize in a timely manner that two crimes are related. 

                We can develop a statistical method to assess the likelihood that two murders are related; we can’t do it today, we can’t do it this month, but we can do it if we work at it.   I know in broad strokes how it can be done.    The first thing we have to do is to measure the similarity between two crimes.    Here’s the order of march:

                1)  Develop a system to measure the similarity of two crimes.

                2)  Write codes to describe several hundred or several thousand murders.

                3)  Refine/modify the coding system based on what you learn from doing this.

                4)  Study murders which are known to be related (committed by the same person or persons) and assess the normal range of similarity scores that describe the relationship between them.

                5)  Study murders which appear to be related but which are not, and look for things which distinguish them.

                6)  Propound a theory for how to convert similarity into a probability of a connection.

                7)  Test your theory.

                8)  Refine your theory.

                9)  Re-test your theory.

                10)  Repeat ad infinitum. 

                Step One can be broken down into parts:

                1)  Develop a coding system to describe a crime.

                2)  Develop a method to estimate the similarity of two crimes, based on their codes.  

                So far in this article, we’ve dealt with the first part of Step One.   In our next article, which will be printed here tomorrow, we will embark into the Second Part of Step One. 

                This is the full list of Cause of Death/Instrument of Death Codes.   It’s pretty ghastly. .. .